Network Computingwww.networkcomputing.com

As IP networks grow, many organizations will find that IP-to-ATM interconnection services are essential, whether for links to an Internet-based VPN (virtual private network) or private IP WAN service. Nearly all the ILECs and most IXCs offer PVCs (Permanent Virtual Circuits) to the Internet, though providing firewall services on a single PVC within what is otherwise a private interface may require some thought, and possibly additional hardware, on the customer's part.

These factors make ATM more affordable for medium-sized WANs, but new low-cost services operating at T1 speeds are opening ATM to an even wider audience. In general, pricing from the major IXCs is close to parity with that of frame relay for similar bandwidth, with higher rates for better classes of service. MCI WorldCom offers identical pricing for the same bandwidth, while Qwest Communications offers ATM PVCs at a lower cost than the frame relay equivalent. ILECs tend to price ATM at a 10 percent to 30 percent premium above frame relay, but most say they are shooting for price parity within the next two years.

ATM's well-noted classes of service (see "ATM Service Classes Explained," page 58) are also driving interest in ATM. These service classes create multiple virtual networks out of a single infrastructure, providing the appropriate level of service for voice, video and various types of data. ATM is very good at controlling delay and variations in delay (or jitter), has superior traffic management, fast rerouting and better mechanisms to control congestion.

Enterprise networks have been slow to adopt ATM, with each carrier generally reporting customer counts in the tens or hundreds, though many expect the totals to double annually in the years ahead. In contrast, frame relay has a huge installed base--estimated at 57,000 customers by Distributed Networking Associates (www.webtorials. com). But nearly all of those frame relay networks have fewer than 50 sites, and low speeds (56/64 Kbps) predominate.

Historically, high-speed services have been beyond the reach of most enterprise customers, and some areas have a shortage of local DS3 loops. The availability of DS3 has improved over the past year, often accompanied by price cuts relative to T1. Bell Atlantic, for example, can offer DS3 for five times the price of T1, whereas the rule of thumb has been 8-to-1. And because OC-3 represents three times the bandwidth of DS3, many customers are moving to OC-3 directly. Most carriers surveyed reported OC-3 pricing at just 1.2-to-1.5 times that of DS3. More than half of GTE's ATM ports run at OC-3.

GTE, Pacific Bell and Southwestern Bell are the only ILECs to roll out n-by-T1 IMA (Inverse Multiplexing for ATM) services, though most of the rest are planning support within a year. In contrast, nearly all the IXCs offer IMA services. The ILECs generally predicted growth to be concentrated at the ends of the spectrum--T1 and OC-3. One Bell Atlantic official we spoke with saw no economic incentive for T1-based IMA, citing the company's low DS3 pricing much more attractive.

Another factor in favor of jumping directly to DS3 or fractional DS3 is that it's more well-suited to real-time traffic. The relatively low throughput of T1 circuits, and the additional latency in the inverse multiplexing process, makes one, faster pipe more attractive for high-volume voice traffic.

Meanwhile, T1-based service's growth may be slowed by the rollout of ADSL (Asymmetric Digital Subscriber Line) services. Many DSLAMs (DSL Access Multiplexers) are ATM-based, and the customer interface is also ATM. Because ADSL services offer comparable bandwidth for a fraction of the cost of T1, small- and branch-office business services may prefer to adopt ATM over ADSL instead of using T1 access. However, local-loop distance limitations in ADSL will keep carriers from achieving T1's reach.

ATM SVC (Switched Virtual Circuit) service offerings are still few and far between. Unfortunately, many of the best applications found in next-generation equipment rely on SVCs. Service providers acknowledged a strong interest among customers in SVCs for voice switching, videoconferencing and other on-demand applications. Yet the challenges with SVC scalability (particularly noted from carriers running the Newbridge platform), along with the difficulty in billing the service, keep them from getting out of trials. Of those carriers with SVCs, Sprint and Qwest bill based on the number of cells or bytes transmitted; only AT&T bases SVC charges on connect time.

Examining the switch platforms that these carriers deploy reveals a lot about where their networks are headed. More than half have standardized on the Ascend CBX-500, which offers trunks at OC-12, or the GX-550, which can trunk at OC-48. But it's the cost of switch ports that really reflects ATM's course: By early 2000, Ascend's DS3 ATM and frame relay ports will be the same price. Further, the bandwidth cost of deploying OC-3 ports is already significantly less than the cost of deploying DS3. The CBX-500 won't offer IMA directly until early 2000, so telcos have to front-end their switches with off-board IMA equipment, which drives up deployment costs.

In contrast, the Newbridge 36170 switches favored by IXC, Cable & Wireless, Pacific Bell and Southwestern Bell offer IMA services directly at the switch, with OC-12 trunking. MCI WorldCom is leaving the Newbridge platform for the Cisco BPX 8600, joining U S West as a customer.

ATM public network SLAs (service-level agreements), CNM (customer network management) platforms and managed services are far from mature. Most carriers don't yet offer ATM-based voice or video, and some don't even offer managed data. Service levels depend upon the class of service used, so it's more complex for customers to verify what they're getting--and no ATM WAN service-level management systems operate below the DS3 level.